Policy Research Working Paper 6003

Does India’s Employment Guarantee Scheme Guarantee Employment?

Puja DuttaRinku Murgai

Martin RavallionDominique van de Walle

The World BankDevelopment Research GroupSouth Asia Social Protection UnitSouth Asia Economic Policy and Poverty UnitHuman Development NetworkMarch 2012

WPS6003P

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

edP

ublic

Dis

clos

ure

Aut

horiz

ed

Produced by the Research Support Team

Abstract

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.

Policy Research Working Paper 6003

In 2005 India introduced an ambitious national anti-poverty program, now called the Mahatma Gandhi National Rural Employment Guarantee Scheme. The program offers up to 100 days of unskilled manual labor per year on public works projects for any rural household member who wants such work at the stipulated minimum wage rate. The aim is to dramatically reduce poverty by providing extra earnings for poor families, as well as empowerment and insurance. If the program worked in practice the way it is designed, then anyone who wanted work on the scheme would get it. However, analysis of data from India’s National Sample Survey

This paper is a joint product of the Director’s Office, Development Research Group; South Asia Social Protection Unit; South Asia Economic Policy and Poverty Unit; Social Protection and Labor Unit, Human Development Network. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at mravallion@worldbank.org.

for 2009/10 reveals considerable un-met demand for work in all states. The authors confirm expectations that poorer families tend to have more demand for work on the scheme, and that (despite the un-met demand) the self-targeting mechanism allows it to reach relatively poor families and backward castes. The extent of the un-met demand is greater in the poorest states—ironically where the scheme is needed most. Labor-market responses to the scheme are likely to be weak. The scheme is attracting poor women into the workforce, although the local-level rationing processes favor men.

Does India‘s Employment Guarantee Scheme

Guarantee Employment?

Puja Dutta, Rinku Murgai, Martin Ravallion and Dominique van de Walle

1

World Bank

JEL: H53, I38

Keywords: workfare, rural poverty, employment, targeting, women

Sector: POV

1 We would like to thank Maria Mini Jos for her very able research assistance. These are the views of the authors

and do not necessarily represent those of the World Bank or of any of its member countries. The authors are grateful

to Emanuela Galasso, Pablo Gottret, Ghazala Mansuri and Giovanna Prennushi for comments.

2

1. Introduction

In 2006, India embarked on an ambitious attempt to fight rural poverty. The National

Rural Employment Guarantee Act of 2005 created a justiciable ―right to work‖ for all households

in rural India through the National Rural Employment Guarantee Scheme, renamed the Mahatma

Gandhi National Rural Employment Guarantee Scheme (MGNREGS) in 2009. This promises

100 days of work per year to all rural households whose adults are willing to do unskilled

manual labor at the statutory minimum wage notified for the program. Work is to be made

available to anyone who asks for it within 15 days of receiving an application to work, failing

which the state government is liable to pay an unemployment allowance. Open village meetings

(Gram Sabhas) are supposed to identify suitable projects and local government institutions

(Gram Panchayats) are given a central role in planning and implementation.

There are a number of distinct ways in which such a scheme tries to reduce poverty. The

most direct and obvious way is by providing extra employment and income to the poorest in

rural areas. The long-standing incentive argument is that the work requirements entail that the

scheme will be ―self-targeting‖ in that the non-poor will not want to do such work, and also

prevents dependency as poor people will readily turn away from the scheme when better

opportunities arise.2

Furthermore, by linking the wage rate for such work to the statutory minimum wage rate,

and guaranteeing work at that wage rate, such a scheme is essentially a means of enforcing that

minimum wage rate on all casual work, including that not covered by the scheme. Indeed, the

existence of such a program can radically alter the bargaining power of poor men and women in

the labor market, and also poor people living in not-so-poor families, by increasing the

reservation wage (the fall-back position if a bargain is not struck). They may then benefit even if

they do not in fact participate in the program.

A scheme such as this can also provide valuable insurance against the many risks faced

by India‘s rural poor in their daily lives. Even those who do not normally need such work can

2 On the incentive arguments for workfare schemes see Besley and Coate (1992).

3

benefit from knowing it is available. This can help underpin otherwise risky investments. And

the gains to the poor can also come with efficiency gains given existing labor market distortions.3

The scheme also tries to address some of the causes of poverty in rural India.4 By its

―bottom-up,‖ demand-driven nature, it aims to empower the rural poor to help them take actions

in various domains that help them escape poverty. It would be naïve to think that such

empowerment will emerge overnight amongst poor people who have faced a history of exclusion

from the processes of public action, and of subjugation to the will of local elites. However,

creating the legal right is certainly a first, positive, step.

The idea of an ―employment guarantee‖ is clearly important to realizing the full benefits

of such a scheme. The gains depend heavily on the scheme‘s ability to accommodate the supply

of work to the demand. That is not going to be easy, given that it requires an open-ended public

spending commitment; similarly to an insurance company, the government must pay up when

shocks hit. This kind of uncertainty about disbursements in risky environments would be a

challenge for any government at any level of economic development.

If the maximum level of spending on the scheme by the center is exogenously fixed for

budget planning purposes then rationing may well be unavoidable at any socially acceptable

wage rate. Or, to put the point slightly differently, the implied wage rate—given the supply of

labor to the scheme and the budget—may be too low to be socially acceptable, with rationing

deemed (implicitly) to be the preferred outcome.5

Even if flexibility in spending is not an issue, accommodating supply to demand could

still be a challenge, particularly in poor areas. Here it should be noted that the provisions of the

Act do not imply that there will be zero cost to the local (State or lower-level) governments when

employing workers under MGNREGS. The center covers a large share of the cost. 6

However,

3 The distortions could be due to monopsony power in rural labor markets (Basu et al., 2009) or labor-tying (Basu,

2011). Nor does the distortion need to be in the rural labor market; it could also be in the urban labor market,

generating excess migration to urban areas (Ravallion, 1990). 4 The scheme also tries to reduce future poverty by creating useful assets. This is not an issue we address here.

5 The policy choice between limited coverage at a socially acceptable (―living‖) wage and wide coverage is studied

in Ravallion (1991). 6 The Central Government bears 90 percent of all variable costs. This includes wage costs and three-quarters of the

non-wage component (working on an assumed 60:40 labor capital ratio). The Centre also provides an additional 6%

of program costs to the States to defray the costs of administering the scheme. States are responsible for paying

4

there are still (relatively skilled) labor requirements at the local level in organizing projects and

workers.7 This burden may well be higher in poor areas, making it harder to afford and

implement such a complex scheme.

This paper examines the performance thus far of MGNREGS in meeting the demand for

work across states. We examine the evidence for India as a whole using the household-level data

from the National Sample Survey for 2009/10. We also use these data to understand who gets

rationed and how this affects the scheme‘s ability to reach India‘s rural poor and other identity-

based groups, notably backward castes, tribes and women. We also discuss the role played by

wage setting on the scheme, and how rationing might be influencing labor market responses.

Finally, we take a closer look at women‘s participation and how this is influenced by the

rationing of work under MGNREGS.

2. Performance in meeting the demand for work across states

The participation rate (P) in MGNREGS can be defined as the proportion of rural

households who obtain work on the scheme. This can be thought of as the product of the

―demand rate‖ (D)—defined as the proportion of rural households who want work on the

scheme—and one minus the ―rationing rate‖ (R)—defined as the proportion amongst those who

wanted work who did not get it. Thus for state i we have the following identity:

iii DRP )1( (1)

Notice that the share of households who are rationed is the product of the rationing rate and the

demand rate. We shall call this the excess demand ( iiiii PDDRED ).

As noted, if MGNREGS worked in practice the way it is designed there would be no

unmet demand for work. This is, of course, an exacting standard. In practice there may be

frictions in implementation leading to some unmet demand, such that those wanting work do not

unemployment allowances from their own budget. 7 The scarcest manpower resource locally is the junior engineer or panchayat technical assistant who can prepare

technical estimates and draw up engineering plans for the works. Andhra Pradesh (AP) has used information

technology to reduce the need for such skilled local staff, by developing standardized and computerized templates

for the engineering plans of common types of works. AP is the exception however. Most other states face shortages

of such skilled local staff.

5

get it in a timely manner. The rationing rate will depend in part on how effective the scheme‘s

implementation is at quickly responding to demand.

How can we measure the true demand for work and hence the rationing rate? The

administrative data indicate virtually no un-met demand for work on MGNREGS. According to

the administrative data, 52.865 million households in India demanded work in 2009/10, and

99.4% (52.53 million) were provided work.8 However, this is deceptive. What is called ―demand

for work‖ in the administrative data is unlikely to reflect the true demand. Several studies have

found that the work application process and the system for recording demand for work is not yet

in place.9 Further, state and local governments have an incentive not to report un-met demand

given that this implies they should pay unemployment allowances. Also, some people will

undoubtedly be deterred from formally obtaining job-cards, demanding work from the officials,

or do not even know that they have the right to make such demands.10

A better measure of demand for work is obtained by asking people directly in their homes

and independently of the scheme. The data we use here comes from the 66th

Round of the NSS

for 2009/10 which included questions on participation and demand for work in MGNREGS that

allow us to estimate demand and rationing rates across states. The survey was conducted

between July 2009 and June 2010 in all states. The Employment-Unemployment Survey

(―Schedule 10.0‖) included three questions on the program: (i) whether the household has a job

card; (ii) whether it got work on the scheme during the last 365 days, for which responses were

coded under three options: got work, sought but did not get work, and did not seek work in

MGNREGS; and (iii) if the household got work, the number of days of work, and the mode of

payment. In addition, the daily status block in Schedule 10 collected information on activities for

all household members during the week preceding the survey, including number of days worked

and wages received, if the respondent worked on MGNREGS public works (PW).

Note that in this paper we limit our definition of participation and rationing to whether

households got work or did not get work. Unmet demand can also take the form of fewer days of

work than desired. Many households who participated were no doubt rationed in that they would

8 Data is from the official Government of India website for MGNREGS (http:\\nrega.nic.in).

9 This is described in a number of the papers in Khera (2011).

10 In Dutta et al. (2012) we provide supportive evidence on this point for Bihar, based on our surveys in 2009/10.

6

have liked more days of work and still had fewer than the 100 days stipulated by the Act. We

have no choice but to ignore this aspect of the scheme‘s performance since the National Sample

Survey (NSS) did not ask how many more days of work the household wanted; all we know is

whether the household wanted work on the scheme.

Table 1 gives the participation rate, the demand rate and the rationing rate across states,

as well as the rural headcount index (H)—the proportion of the population living below our

updated versions of the Planning Commission‘s latest poverty lines. (The table also gives the

female share of employment, to which we return later.) ―Demand‖ is defined as either getting

work on the scheme or seeking work but not getting it. For India as a whole, 45% of rural

households wanted work on the scheme. Of these, 56% got work—a national rationing rate of

44%. The rationing rate varied from 15% in Rajasthan to 84% in the Punjab. Only three states

have rationing rates under 20%. There is clearly a large excess demand for work.

A striking observation about the data in Table 1 is that participation rates are only weakly

correlated with the rural headcount index across states, as can be seen in Figure 1. If MGNREGS

worked the way the Act intended then this weak correlation would be surprising, as one would

expect the scheme to be more attractive to poor people, and hence have higher take up in poorer

states. The same point holds for public spending on MGNREGS. Table 2 gives summary

statistics on spending per capita for 2009/10 and 2010/11. The correlation between MGNREGS

spending per capita and the poverty rate is -0.02 using spending in 2009/10 and 0.04 for 2010/11.

Why is MGNREGS not more active in poorer states? We postulate that being a poor state

has two opposing effects on participation. First, there is an effect of greater poverty via a higher

demand for MGNREGS work. Call this the ―demand effect‖ of poverty. We see confirmation of

this in Figure 2, which shows the expected positive correlation between D and H (r=0.50). The

second effect is that poorer states tend to have greater un-met demand for work on the scheme.

Call this the ―rationing effect.‖ We suggest three reasons why the rationing effect would work in

the opposite direction to the demand effect. First, poorer states will be less able to afford the

share of the costs that are borne by the state and local governments. Second, poorer states will

tend to have weaker capacity for administering such a scheme. Third, the poor may well be less

empowered in poorer states. As we will see in the next section, both poor and non-poor people

have a demand for work on the scheme, though the demand is greater amongst the poor. If poor

7

people tend to have less power to influence local decision making (reflected in lower awareness

of their rights under the Act), then a higher poverty rate will lead the state government to put less

weight on the need to accommodate the demand for work.11

We cannot say anything from these data about the relative importance of these three

factors to the overall rationing effect. But we can certainly confirm that there is evidence of a

rationing effect. Poorer states have greater unmet demand for MGNREGS, as can be seen in

Figure 3, which plots the excess demand—i.e., the share of the rural population that is rationed,

as given by the rationing rate times the demand rate—against the poverty rate. However, it

should be also noted that there is variation among poorer states. Some of the poorest states

(Bihar, Jharkhand and Orissa) have low participation rates and high levels of unmet demand.

This is in contrast to other poor states like Chhattisgarh, Rajasthan, Madhya Pradesh and West

Bengal that perform better in providing employment under the scheme. For example, at a similar

poverty rate, Chhattisgarh has a participation rate almost five times that of Bihar. Public

spending is also lower in Bihar at roughly one-third of the level in Chhattisgarh.

Using the identity in equation (1), we can do a simple regression decomposition for

identifying these two effects in the data by writing:

D

ii

DD

i HD (2.1)

ED

ii

EDED

i HED (2.2)

(Here k

i for k=D, ED are zero-mean regression error terms and kk , are parameters.) Thus

equation (2.1) minus (2.2) gives how the overall participation rate varies with H. The regression

coefficient of demand for MGNREGS (based on the NSS responses) on the state poverty rate is

0.583 (st. error=0.189), meaning that a ten percentage point increase in the poverty rate comes

with about a 6 percentage point increase in the share of rural households demanding MGNREGS

work, on average. The regression coefficient of ED on H is 0.434 (st.error=0.097). The net effect

(the estimate ofEDD ) is 0.149, but it is not significantly different from zero

(st.error=0.293). Statistically, the two opposing effects can be said to cancel each other out,

11

Note that this third reason for the direct effect of poverty is not consistent with a model of public decision making

based on standard utilitarian calculus. For then one would expect the policy weight on accommodating the demand

for work to be higher in states with a higher share of poor people who need that work more than the non-poor.

8

giving the relationship in Figure 1, whereby poorer states have no higher participation in

MGNREGS, despite the greater demand for work on the scheme.

To better understand this strong rationing effect of poverty, in Dutta et al. (2012) we

study more closely the performance of the state of India with the highest poverty rate, Bihar.

Drawing on various (qualitative and quantitative) data sources, including our own special-

purpose surveys, we argue that low administrative capacity for implementing the scheme and

weak empowerment of poor people are at work in Bihar. Furthermore, we argue that changing

one alone will not assure that MGNREGS will reach its potential in India‘s poorest areas.

Effective action on both fronts will be necessary,

3. Is rationing undermining the self-targeting mechanism?

By insisting that participants do physically demanding manual work at a low wage rate,

workfare schemes such as MGNREGS aim to be self-targeted, in that non-poor people will not

want to participate. The substantial rationing that we have demonstrated above raises the

question of how well this self-targeting mechanism works in practice. The fact that there is

rationing does not mean that targeting will not be pro-poor. For one thing, the manual work

requirement at a low wage rate will still discourage non-poor people from wanting to participate.

For another, the local authorities doing the rationing may well favor the poor. The local officials

who are deciding who gets work could either enhance or diminish the scheme‘s targeting

performance. There has been no comprehensive national assessment of targeting performance.

The quantitative studies that have been done so far have been based on selected samples and the

tests used have often been problematic.12

What does the evidence from the NSS survey for

2009/10 suggest?

12

Jha et al. (2009) report evidence that households with larger landholdings were more likely to participate in the

scheme in Andhra Pradesh (AP), though they find evidence of better targeting in Rajasthan. They conclude that the

scheme is being ―captured‖ by the non poor in AP. Note, however, that their regressions control for other variables

that may well capture poverty, including occupation and whether the household has a BPL card. The full regressions

are not presented in their paper, but it may well be that having a BPL card (say) is already capturing the pro-poor

targeting of MGNREGS, but that the BPL card puts too high a weight on landlessness from the point of view of

explaining participation. Then the amount of land may appear to have the wrong sign, even though the scheme is

targeting the poor. By contrast, the results of Liu and Deininger (2010) suggest quite pro-poor targeting of

MGNREGS in AP. Shariff (2009) reports participation regressions for MGNREGS in selected backward districts of

northern states (including some districts in Bihar). Some of the regression coefficients also suggest perverse

targeting. Shariff is careful in interpreting the results though the same inferential concerns hold as for the study by

9

Table 3 gives the participation rate, demand rate and the rationing rate by rural household

quintiles defined on household consumption per person from the survey.13

As expected, we see

that demand for work on MGNREGS declines with consumption per person. Richer households

are less likely to want to do this work, although there is demand even amongst the richest

quintile in rural areas. Consistently with the incidence of expressed demand, we also see that the

proportion of households who have obtained job cards declines with consumption per person.

But notice that the demand rate is higher than the proportion with job cards; there are many

households who express demand for work who have not obtained job cards.

Across India as a whole, the rationing rate tends to rise with consumption per person. The

local-level processes of deciding who gets work amongst those who want it entails that poorer

households are less likely to be rationed, although the difference is modest. Thus we see that the

participation rate declines with household consumption per person even more steeply in

proportionate terms than the demand rate.14

By contrast the absolute gap between the

participation rates for the poorest quintile and the richest is attenuated by rationing, given that its

overall incidence (the excess demand rate) falls with consumption (Table 3).

Quintile averages lose a lot of detail. Figure 4 provides a finer representation of the data

using the household-level data from the NSS. Panel (a) gives the non-parametric regression

function—a locally smoothed scatter plot—of the participation rate against consumption per

person, with the latter converted into ranks and normalized to be between 0 and 100. Thus the

horizontal axis gives the percentile of the consumption distribution.15

The lower panel (b) gives

the rationing rate. (The demand rate shows a similar pattern to panel (a), though at higher levels

given the rationing.) In both cases we also give the regression functions with controls for state

effects, adding dummy variables for the state and using the PLREG procedure of Lokshin

(2006).

Jha et al.. 13

Household quintiles were drawn after correcting per capita consumption for cost-of-living differences across

states using the price deflators implicit in the Tendulkar poverty lines. 14

The participation rate for the richest quintile is a lower proportion of that for the poorest than is the case for the

corresponding demand rates. 15

Alternatively one can just use (say) log consumption per person. However, given the uneven spread of the data

across levels of consumption, this can be deceptive. Using percentiles instead assures a uniform distribution of the

data.

10

We see that the participation rate declines rather slowly until one reaches about the 50th

percentile of the rural distribution. In fact there is no perceptible decline in participation in a

neighborhood of the national headcount index of poverty of 36%; households just below the

official poverty line are no more likely to participate in MGNREGS than those just above the

line. The marked decline in participation rates does not emerge until we get to the upper half of

the rural consumption distribution. By the 90th

percentile, the participation rate reaches about

10%. Although far fewer ―rich‖ rural households participate, there are still some. This could

reflect recent shocks, or poor individuals within generally well-off households.

The rationing rate follows a U-shape, declining initially as consumption increases until

one reaches about the median, but then rising. However, this clearly stems from the high

rationing rates in poor states that we have demonstrated above. Thus we see that, when we add

state fixed effects, a steady increase in the rationing rate emerges as consumption increases

across the whole distribution (Figure 4(b)).

The fact that the rationing rate tends to be lower for the poor does not, however, imply

that more rationing would improve targeting. What the numbers in Table 3 and Figure 4(b)

reflect is the rationing process at a given level of participation. When the participation rate rises

through a reduction in rationing the self-targeting mechanism will start to play a bigger role. We

will see evidence of this when we compare targeting performance across states with very

different participation rates.

Also notice that, amongst participants, the days of work received shows a slightly

positive gradient with consumption per person. The pro-poor targeting is achieved through both

demand for work and the rationing of work, not by the amount of work actually received.

It is of interest to compare targeting performance across states. There are many measures

of ―targeting performance‖ in the literature that might be used for this purpose. Ravallion (2009)

surveys the various measures and tests their performance in predicting the impacts on poverty of

a large antipoverty program in China, called the Di Bao program. (This provides cash transfers

targeted to those with income below the locally-determined Di Bao poverty lines.) Amongst all

standard targeting measures, the one that performed the best (and by a wide margin) in predicting

the program‘s impact on poverty was the ―targeting differential‖ (TD), originally proposed by

11

Ravallion (2000). The reason is that TD better reflects differences in coverage—the proportion

of the poor receiving the program—than other standard measures, which focus more on how well

the scheme avoids leakage—the proportion of the non-poor receiving the program. This is likely

to hold for other programs besides Di Bao.

In the present context, this can be defined as the difference between the MGNREGS

participation rate for the poor and that for the non-poor. In obvious notation:

poornon

i

poor

ii PPTD (3)

Here Pi (i=poor, non-poor) is again the participation rate (as defined by equation (1)), but this

time differentiated between the poor and non-poor. To interpret the targeting differential, note

that when only poor people get help from the program and all of them are covered, TD = 1,

which is the measure‘s upper bound; when only the non-poor get the program and all of them do,

TD = -1, its lower bound. This measure is easy to interpret, and it automatically reflects both

leakage to the non-poor and coverage of the poor.

Table 4 gives the TD and participation rates for the poor and non-poor.16

Participation

rates among the poor vary enormously across states, from a low of 0.10 in Maharashtra to a high

of 0.73 in Rajasthan. They also vary a lot among the non-poor. Although participation rates are

always higher for the poor, the gap with that for the non-poor is not large. The targeting

differential for India as a whole is 0.12. (The TD for China‘s Di Bao program mentioned above

was 0.27.) Madhya Pradesh has the highest TD, at 0.22, while Kerala has the lowest, at 0.01.

Table 4 gives the rationing rates for the poor and non-poor. Consistently with the all-

India results in Table 3, we see that the non-poor are rationed more than the poor in almost all

states (the only exceptions are Kerala and Rajasthan).

The TD is determined by how the demand rates and the rationing rates vary between the

poor and non-poor. We can use a simple decomposition method to show how much of the TD is

due to each factor:

16

To divide the population into poor and non-poor, we use poverty lines that deliver the same poverty rates using

the abridged consumption module in the Employment Schedule, as those that are obtained from the Consumption

schedule, and reported in Table 1.

12

residualRRDDDRTD poornon

i

poor

i

poornon

i

poor

ii )())(1( (4)

(―Self-targeting effect‖) (―Rationing effect‖)

Here the bars denote fixed reference values, while Di and R

i are the demand rates and rationing

rates for i=poor, non-poor, respectively. TD can thus be interpreted as the ―self-targeting effect‖

(greater demand for work amongst the poor) net of the rationing effect (the extent to which the

poor might be rationed more). (Since the decomposition is not exact—given the nonlinearity in

equation (1)—there is also a residual.)

Applying this decomposition, and using the all-India values for the reference we find that

85.6% of the national TD is attributable to the difference in demand between the poor and non-

poor while 13.7% is due to the difference in rationing rates. (The residual is negligible.) There

are differences across states, though the demand effect dominates in 17 of the 20 states. So,

despite the rationing, the bulk of the pro-poor targeting is coming through the self-targeting

mechanism.

Targeting performance is better in states with higher overall participation rates. Figure 5

plots the two participation rates from Table 4 against the overall participation rate (Table 1). We

see that the TD—the gap between the two lines—rises with the overall participation rate, and the

two are strongly correlated (r=0.748).

Targeting performance also tends to be worse in the states with higher levels of rationing

(the correlation between TD and rationing rate is -0.71). However, this arises because overall

participation rates are low in states with higher degrees of rationing. Indeed, once one controls

for the participation rate, there is no significant partial correlation between the TD and the

rationing rate (the t-statistic is -0.611).17

So we find that higher overall participation rates tend to come with better targeting

performance and lower rationing rates. The fact that targeting performance improves as the

program expands makes this an example of what Lanjouw and Ravallion (1999) call ―early

capture‖ by the non-poor, which they showed to be a common feature of access to safety-nets

17

The TD is also positively correlated with the demand rate (r=0.671), but this too vanishes when one controls for

the participation rate (the t-statistic for the partial correlation coefficient is 0.151).

13

and schooling in India.18

Lanjouw and Ravallion also show in a theoretical model of the political

economy of targeted programs that for programs with relatively large start-up costs, early capture

by the nonpoor may be the only politically feasible option (especially when the start-up costs

must be financed domestically). So this feature of MGNREGS is possibly not surprising.

Targeting by social groups (castes and tribes) is another dimension of interest. Qualitative

studies have suggested that scheduled castes (SC), scheduled tribes (ST) and women—groups

that have traditionally been excluded—have benefitted disproportionately from the scheme.19

We

shall return to discuss participation by women in section 5. Here we focus on the scheme‘s

performance in reaching ST, SC and other backward caste (OBC) households.

Table 5 gives the participation rates by these groupings. Nationally, 42% and 34% of

rural ST and SC households respectively participated. Participation was lower for OBCs at 21%

and lowest for all others, at 16%. But there is a wide range across states. For STs, the range is

from 6% of households participating in Maharashtra to 82% in Rajasthan, while for SC

households it is from 2% to 65%, and for the same states. Figure 6 plots the participation rates

against the overall participation rates across states. Similarly to targeting by poverty status, we

see that the participation rates for ST, SC and OBCs rise faster as the overall participation rates

rise, suggesting that the targeting of disadvantaged castes improves with program expansion.

Table 5 also gives the targeting differential for ST, SC and OBCs together, defined as

their (weighted) average participation rate less the participation rate for ―others.‖ This ―caste

TD‖ varies from -0.02 in Assam to 0.29 in Chhattisgarh, with a national mean of 0.12, almost

identical to the national ―poverty TD‖ in Table 4. Similarly to the poverty TD, the ―caste TD‖ is

positively correlated with the overall participation rates (r=0.723).

4. Wages and rationing on MGNREGS

There have been a number of concerns about the stipulated wage rates for the program.

On the one hand, it is argued that setting scheme wages below the state-mandated rates under the

Minimum Wages Act is a violation of the law and tantamount to ―forced labor,‖ a stand that has

18

Lanjouw and Ravallion (1999) study public works programs, the Integrated Rural Development Program, the

Public Distribution System and school enrolments. They used data for the 1990s. 19

See Drèze and Khera (2009).

14

been recently upheld by the Supreme Court.20

On the other hand, concerns have been raised that

the wage rate on MGNREGS is being set too high, relative to actual casual labor market wages.

The concern here is that the scheme will attract workers from market work and so bid up the

market wage rate.21

(Though to many supporters of the scheme this is counted as a benefit.)

What does the evidence suggest? Table 6 gives the average wage rates from the

administrative data. These are calculated as total MGNREGS spending on unskilled labor

divided by total person days of employment provided.22

The table also reports estimates for

average wages in (private) casual labor from the NSS for the same year.23

We see that it is not the case that the MGNREGS wage rate is everywhere well above the

market wage rate. Indeed, for India as a whole the two wages are quite close. If rural India was

one labor market one might conjecture that the scheme has indeed brought the two wage rates

into parity. However, rural India is not one labor market, as mobility is clearly imperfect. Across

states we see that for half of them the MGNREGS wage rate in 2009/10 is actually lower than

the average wage rate for casual labor (Table 6).

Given the extent of rationing we have reported earlier in this paper, it does not seem

plausible that the scheme would be having a large impact on wages for other casual work, let

alone resulting in a higher casual wage than for MGNREGS in half the states. For example, with

only 17% of those who wanted work on the scheme in Punjab getting that work, it is hard to

believe that the fact that the casual (non-PW) wage rate is above the MGNREGS wage rate is

due to competition with the scheme for workers.

That said, we do find that the relative wage—defined as the mean wage rate for casual

(non-PW) labor divided by the MGNREGS wage—tends to be lower in states with higher levels

of un-met demand, as measured by the difference between the demand rate and the participation

20

The Supreme Court has refused to stay a recent Karnataka High Court verdict that affirms that the Central

Government is liable to pay wages in tandem with the state minimum wage rate. 21

Evidence for this effect is reported by Imbert and Papp (2011) who compare districts that started early on the

scheme with those that started later. However, they do not examine the extent of rationing. 22

Note that the scheme stipulates both piece rates and daily rates. Under the piece rate, whether a given worker can

earn the mandated wage rate depends on her work effort. If the scheme attracts workers with lower than average

physical ability then the realized average wage rate by our calculations can fall short of the mandated wage. 23

Note that the reference periods for MGNREGS and casual market wages reported in the table are slightly different

(see notes to Table 6).

15

rate. Let EGS

ii WW / be the relative wage in state I, where iW is the wage rate in the (non-PW)

casual labor market and EGS

iW is the MGNREGS wage rate. The correlation coefficient between

)/ln( EGS

ii WW and un-met demand ( iED ) is –0.558 which is significant at the 1% level.24

However, there are two reasons to question whether this really reflects greater tightening

of the casual labor market in states where there is less un-met demand for work on the scheme.

First, the implied relative wage rate at zero rationing is too high to be believed. We postulate that

the relative wage depends on the excess demand as follows:25

iii

EGS

ii PDWW )()/ln( (5)

The expected value of the log relative wage rate when all demand for work on MGNEGS has

been satisfied is then given by . Using the data in Table 1 and 5, the estimated value of is

0.398 (st. error=0.105) and the estimate of is –1.777 (st. error=0.519; R

2=0.311.) This implies

that, when all demand for work is satisfied, the market wage rate would be 50% higher than the

MGNREGS wage rate.26

Yet the work is very similar, and there is no obvious reason why such a

differential would exist in equilibrium.

Second, the excess demand variable could well be picking up some other factor

correlated with it. Following our discussion in section 2, poverty is a plausible candidate. Adding

the poverty rate to equation (5), we find that the effect of excess demand becomes insignificant

(prob.=0.19) while the poverty rate is significant (a coefficient of 0.828, with st. error of 0.333).

A higher poverty rate may be associated with greater landlessness and hence a larger supply of

casual labor, bringing down the wage rate.

5. Rationing and the participation of women

Nationally, almost half (48%) of the employment as registered in the administrative data

for 2009/10 goes to women.27

This is very high for a country where a minority of women

24

The fit is slightly better using the log relative wage; using EGS

ii WW / instead the correlation coefficient is -0.520. 25

Using EGS

ii WW / as the dependent variable the estimated intercept is 1.466 (s.e.=0.132). We also tried a quadratic

function of ED but this did not improve the fit. 26

Note that exp(0.398)=1.489 is the implied ratio of the levels of wages. 27

While the administrative data are clearly inadequate for measuring aggregate demand for work, there is no

16

participates in the paid labor force; for example, women‘s participation rate in MGNREGS is

about twice their share of other (non-PW) casual wage work.28

The variation across states is

striking; between the two extremes, only 7% of the work goes to women in Jammu & Kashmir as

compared to 88% in Kerala (Table 1). The female share on MGNREGS work is greater than

their share of the work in the casual wage labor market in all states.

Women are less likely to participate (relative to men) in MGNREGS in poorer states.

Figure 7 plots the share of person days of employment going to women against the poverty rate.

We see a negative correlation (r = –0.47). By contrast, women‘s share of casual (non-PW) wage

work tends to be slightly higher on average in poorer states, though the difference is not

statistically significant (r=0.09). While the scheme is clearly bringing women into the paid

workforce, it is less effective in doing so in states with higher poverty rates.

Why do we see less of the available work going to women in poorer states? The NSS

does not allow us to identify rationing at the individual level (it is a household variable).

However, we can gain some clues to how rationing has affected women from the inter-state

comparisons. It would seem unlikely that the effect of household poverty on demand for work

amongst women is any different than amongst men. Assuming that the effect of being a poorer

state on demand for work is the same for men and women, the pattern in Figure 7 suggests that

the rationing process is less favorable to women in poorer states.

Do women have equal access to the scheme, when they need it? Again, we cannot give a

direct answer from the survey data. However, the patterns in the inter-state data are suggestive.

We observe a negative correlation between the female share of work and the overall rationing

rate (Figure 8). It might be conjectured that this correlation actually reflects differences in the

extent of poverty. Women may well be less aware of their rights and less empowered to demand

work in poorer states. For example, when other work is scarce, they may get crowded out by

men. However, the negative correlation between the female share of work and the rationing rate

persists when we control for the poverty rate, as can be seen in Table 7, which gives regressions

for the female share of employment on MGNREGS against both the rationing rate and the

obvious reason to question their veracity for measuring the gender composition of the work provided. 28

We estimate that the share of women in the total person days of casual labor in 2009/10 was 23.3%, based on the

2009/10 NSS.

17

poverty rate. This negative correlation between the overall rationing rate and the share of work

going to women implies that women are more likely to be rationed than men.

In Dutta et al. (2012) we are able to directly measure rationing at the individual level in

Bihar, using our specially designed surveys. Then we find direct evidence that the rationing rate

is higher for women than men. In the first round of our survey (2008/09) the rationing rate (R)

was 85% for women in Bihar, versus 65% for men. This had narrowed somewhat by the second

round (2009/10) to 85% versus 74%, but only because of greater rationing of men.

Gender differences in the opportunities available in the casual labor market can also be

expected to influence demand for work. We find that the female market wage rate has a

significant negative effect on women‘s share of the work provided, while the male wage rate has

the opposite effect (Table 7).29

The female wage relative to the male wage is the relevant

variable. This suggests that there is an intra-household substitution effect; for example, when

casual labor market opportunities are good for men but bad for women this makes it easier for

women to get the (limited) number of jobs available on the scheme.30

The wage effect is strong

statistically, and greatly increases the explanatory power.31

The negative effect of rationing on

women‘s access to the scheme also persists when we control for differences in the wages

received for private (non-PW) casual work. The final column gives our preferred regression in

which the share of MGNREGS work going to women depends on the overall rationing rate—

implying that women are more rationed than men—and the female market wage relative to the

male wage.

29

We tested an encompassing specification in which the log of the male wage rate, log of the female wage rate and

log of the MGNREGS wage rates entered separately. The homogeneity restriction that the sum of the coefficients

equals zero could not be rejected (F(1,14)=0.41; prob.=0.53), but nor could we reject the null that it was the log of

the female wage relative to the male wage that mattered, with the MGNREGS wage having no affect (F=1.49;

prob.=0.26). Also, the MGNREGS wage rate on its own was not significant. So we opted for a specification in

which it is the log of the relative wage that is the regressor as in Table 7. Table 7 also gives a specification with male

and female wages entering separately. 30

A similar result was found by Datt and Ravallion (1994) in studying time allocation within households in response

to the availability of work under Maharashtra‘s Employment Guarantee Scheme. 31

We also tested for an effect of the female share of other (non-public works) casual labor. This is endogenous but it

allows us to control for local social norms that influence the propensity for women to do any casual wage labor. The

new variable was not, however, significant and the coefficients on the other variables were affected little.

18

6. Conclusions

There has been much public debate about India‘s Mahatma Gandhi National Rural

Employment Guarantee Scheme since it was introduced. There have been many media reports

and some selective surveys, covering at most a few states and/or selected districts. This paper

uses India‘s National Sample Survey of 2009/10 to test some of the claims that have been made

in past debates using data for all the major states of India. We have focused on a distinctive and

important feature of MGNREGS: the guarantee of employment at the stipulated wage rates.

We confirm expectations that the demand for work on MGNREGS tends to be higher in

poorer states. This appears to reflect the scheme‘s built-in ―self-targeting‖ mechanism, whereby

non-poor people find work on the scheme less attractive than do poor people.

However, actual participation rates in the scheme are not (as a rule) any higher in poorer

states where it is needed the most. The reason for this paradox lies in the differences in the extent

to which the employment guarantee is honored. The answer to the question posed in our title is

clearly ―no.‖ Rationing is common, but far more so in some of the poorest states.

We do not find that the local-level processes determining who gets work amongst those

who want it are generally skewed against the poor. There are sure to be places where this is

happening (and qualitative field reports have provided examples). But it does not appear to stand

up as a generalization. We do find evidence that the poor fare somewhat less well when it comes

to the total number of days of work they manage to get on the scheme. However, despite the

pervasive rationing we find, it is plain that the scheme is still reaching poor people and also

reaching the scheduled tribes and backward castes.

Participation rates on the scheme are higher for poor people than others. This holds at the

official poverty line, but the scheme is also reaching many families just above the official line. It

is only at relatively high consumption levels that participation drops off sharply. This should not

be interpreted as indicating that well-off families in rural India are turning to MGREGS. There

may well be shocks that are not evident in the household consumption aggregates. And there

may be individual needs for help that are not evident in those aggregates.

19

Targeting performance varies across states. Some of those living above the official

poverty line in better-off states will no doubt be relatively poor, and need help from the scheme.

The overall participation rate seems to be an important factor in accounting for these inter-state

differences in targeting performance, with the scheme being more pro-poor and reaching

scheduled tribes and backward castes more effectively in states with higher overall participation

rates.

While the allocation of work through the local-level rationing process is not working

against the poor, there are clearly many poor people who are not getting help because the

employment guarantee is not in operation almost anywhere (Himachal Pradesh, Rajasthan and

Tamil Nadu could be counted as the exceptions, where 80% or more of those who want work got

it). And other potential benefits of the scheme to poor people are almost certainly undermined by

the extensive rationing, notably the empowerment gains and the insurance benefits. The first-

order problem for MGNREGS is the level of un-met demand.

While the scheme is clearly popular with women—who have a participation rate that is

double their participation rate in the casual labor market—the rationing process does not appear

to be favoring them. We also find evidence of a strong effect of relative wages on women‘s

participation—both wages on the scheme relative to the market wage and the male-female

differential in market wages. As one would expect, poor families often choose whether it is the

man or the woman who goes to the scheme according to relative wages.

It has been claimed by some observers that the scheme is driving up wages for other

work, such as in agriculture; some observers see this as a good thing, others not. For India as a

whole, we find that the scheme‘s average wage rate was roughly in line with the casual labor

market in 2009/10. This might look like a competitive labor market equilibrium, but that view is

hard to reconcile with the extensive rationing we find. Interestingly, we do find a significant

negative correlation between the extent of rationing and the wage rate in the casual labor market

relative to the wage rate on the scheme. Although this is suggestive, on closer inspection we are

more inclined to think that other economic factors are at work. Indeed, the correlation largely

vanishes when we control for the level of poverty. Poorer states tend to see both more rationing

of work on the scheme and lower casual wages—possibly due to a greater supply of labor given

the extent of rural landlessness.

20

Figure 1: Participation rates in MGNREGS are only weakly correlated with the incidence

of poverty across the states of India

.0

.1

.2

.3

.4

.5

.6

.7

10 15 20 25 30 35 40 45 50 55 60

Headcount index of rural poverty 2009/10

Par

ticip

atio

n ra

te in

MG

NR

EG

S (s

hare

of r

ural

hou

seho

lds)

r=0.13

21

Figure 2: Demand for MGNREGS work is greater in poorer states

Figure 3: Poorer states have greater unmet demand for work on MGNREGS

.1

.2

.3

.4

.5

.6

.7

.8

10 15 20 25 30 35 40 45 50 55 60

Headcount index of rural poverty 2009/10

Dem

and

rate

: Sha

re o

f rur

al h

ouse

hold

s w

ho w

ant w

ork

r=0.50

.05

.10

.15

.20

.25

.30

.35

.40

10 15 20 25 30 35 40 45 50 55 60

Headcount index of rural poverty 2009/10

Shar

e of

rura

l hou

seho

lds

who

wer

e ra

tione

d Biharr=0.74

Jharkhand

Orissa

Punjab

Himachal PradeshTamil Nadu

RajasthanKerala

Chhattisgarh

22

Figure 4: Participation, demand and rationing by consumption per person, rural India

(a) Participation rate

(b) Rationing rate

0.1

.2.3

.4

Par

ticip

atio

n ra

te in

MG

NR

EG

S

(pro

porti

on o

f pop

ulat

ion

parti

cipa

ting)

0 20 40 60 80 100Percentile ranked by consumption per person

(1=poorest; 100=richest)

LOWESSPLREG WITH STATE CTRL

Rural IndiaParticipation rate in MGNREGS as a function of consumption

.3.4

.5.6

.7

Rat

ioni

ng ra

te o

n M

GN

RE

GS

(sha

re o

f tho

se w

antin

g w

ork

who

did

not

get

it)

0 20 40 60 80 100Percentile ranked by consumption per person

(1=poorest; 100=richest)

LOWESSPLREG WITH STATE CTRL

Rural IndiaRationing rate on MGNREGS as a function of consumption

23

Figure 5: The participation rate of poor households in MGNREGS rises faster than that for

the nonpoor as the overall participation rate increases

Figure 6: Participation rates for ST, SC and OBC in MGNREGS rise faster than that for

‘other’ castes as the overall participation rate increases

.0

.1

.2

.3

.4

.5

.6

.7

.8

.0 .1 .2 .3 .4 .5 .6 .7 .8Overall participation rate

Parti

cipa

tion

rate

for p

oor/n

onpo

or

Poor

Nonpoor

.0

.1

.2

.3

.4

.5

.6

.7

.8

.9

.0 .1 .2 .3 .4 .5 .6 .7 .8 .9

STSCOBCOther

ScheduledTribe

ScheduledCaste

OtherBackwardCastes

Other

Overall participation rate

Parti

cipa

tion

rate

for e

ach

grou

p

24

Figure 7: The share of work going to women tends to be lower in poorer states

Figure 8: Higher rationing tends to come with a lower share of the work going to women

0.0

0.2

0.4

0.6

0.8

1.0

0 5 10 15 20 25 30 35 40 45 50 55 60

Headcount index of rural poverty 2009/10 (% below poverty line)

Female share of non-PW workFemale share of MGNREGS

Fem

ale

shar

e of

wor

k

Female share ofMGNREGS

Female share ofnon-PW work

0

20

40

60

80

100

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Rationing rate (share of those who wanted work who get it)

Fem

ale

shar

e of

per

son

days

of e

mpl

oym

ent (

%)

KeralaTamil Nadu

Jammu & Kashmir

r=-0.58

25

Table 1: Summary statistics 2009/10

State

Headcount

index of

poverty (% below

poverty line)

Participation

rate (share of

rural h‘holds

working on

MGNREGS)

Demand rate

(share of rural

h‘holds who

want work on

MGNREGS)

Rationing rate

(share of rural

h‘holds who

wanted work but

did not get it)

Female share

of employment

on MGNREGS (% of total

person days)

Andhra Pradesh 20.64 0.354 0.472 0.249 58.1

Assam 42.28 0.182 0.413 0.559 27.7

Bihar 56.47 0.099 0.461 0.785 30.0

Chhattisgarh 56.39 0.479 0.690 0.306 49.2

Gujarat 32.54 0.215 0.382 0.438 47.5

Haryana 24.18 0.051 0.195 0.738 35.6

Himachal Pradesh 11.90 0.334 0.418 0.202 46.0

Jammu & Kashmir n.a. 0.097 0.334 0.709 7.0

Jharkhand 43.50 0.192 0.517 0.628 34.3

Karnataka 31.34 0.080 0.228 0.648 36.8

Kerala 11.74 0.112 0.232 0.517 88.2

Madhya Pradesh 45.85 0.406 0.646 0.371 44.3

Maharashtra 33.90 0.044 0.277 0.840 39.8

Orissa 49.93 0.220 0.507 0.567 36.3

Punjab 19.44 0.052 0.312 0.833 26.0

Rajasthan 31.2 0.618 0.732 0.155 66.9

Tamil Nadu 22.81 0.335 0.414 0.190 82.9

Uttar Pradesh 40.75 0.162 0.350 0.536 21.7

Uttarakhand n.a. 0.292 0.406 0.280 40.1

West Bengal 35.05 0.432 0.658 0.344 33.4

All India 36.43 0.249 0.447 0.444 48.1

Notes and sources: Poverty rates are based on Tendulkar poverty lines updated from 2004/05 to 2009/10 using state-

specific consumer price indices for agricultural laborers (CPIAL) and per capita consumption expenditures in

Schedule 1.0. Poverty rates for J&K and Uttarakhand not reported because data not available on state-level CPIAL.

Female share of person days from MGNREGS administrative data (http:\\nrega.nic.in). Remaining columns from

authors‘ calculations from unit record data of 2009/10 National Sample Survey Schedule 1 (for headcount rate) and

Schedule 10.

26

Table 2: Program expenditures per capita across states

Expenditure per capita (Rs) 2009/10 2010/11 Andhra Pradesh 749 896 Assam 406 358 Bihar 214 309 Chhattisgarh 723 884 Gujarat 214 226 Haryana 87 128 Himachal Pradesh 936 837 Jammu and Kashmir 221 445 Jharkhand 586 539 Karnataka 742 683 Kerala 186 276 Madhya Pradesh 734 707 Maharashtra 54 60 Orissa 281 455 Punjab 89 98 Rajasthan 1133 647 Tamil Nadu 555 744 Uttar Pradesh 389 365 Uttarakhand 406 539 West Bengal 335 399 All India 464 477 Notes and sources: Cumulative expenditures (including wage and non-wage spending) in current prices during the

2009/10 and 2010/11 FY were obtained from the Ministry of Rural Development website (http:\\nrega.nic.in). To

calculate expenditure per capita the authors used the population projections for 2009 and 2010 done by Registrar

General of India.

Table 3: Coverage of MGNREGS across consumption quintiles of the rural population of

India, 2009/10

Quintiles Participation

rate Demand

rate Rationing

rate Excess

demand rate

Share of

HHs with

jobcard

Mean person

days amongst

participating

HHs

Mean

person days

among all

rural HHs Q1 (Poorest) 0.335 0.609 0.450 0.274 0.465 33.7 11.3 Q2 0.297 0.540 0.450 0.243 0.414 36.2 10.7 Q3 0.273 0.507 0.462 0.234 0.385 38.3 10.4 Q4 0.226 0.434 0.479 0.208 0.329 40.0 9.0 Q5 (Richest) 0.138 0.309 0.553 0.171 0.218 40.0 5.5 All 0.242 0.462 0.476 0.220 0.347 37.4 9.0 Notes: The participation rate is the share of rural households working on MGNREGS. The demand rate is the

share of rural households who want work on the program. The rationing rate is the share of those who wanted work

who did not get it. Source: Authors‘ estimates from NSS (2009/10).

27

Table 4: Targeting performance of MGNREGS across states

State

Participation

rate for the

poor

Participation

rate for the

non-poor

Targeting

differential

Rationing

rate for the

poor

Rationing

rate for the

non-poor

Andhra Pradesh 0.513 0.322 0.191 0.215 0.259 Assam 0.233 0.149 0.085 0.523 0.590 Bihar 0.127 0.075 0.052 0.756 0.816 Chhattisgarh 0.571 0.386 0.186 0.260 0.366 Gujarat 0.298 0.185 0.114 0.319 0.490 Haryana 0.106 0.037 0.069 0.701 0.760 Himachal Pradesh 0.510 0.318 0.192 0.173 0.206 Jharkhand 0.237 0.163 0.075 0.613 0.641 Karnataka 0.126 0.065 0.061 0.503 0.703 Kerala 0.116 0.112 0.005 0.535 0.516 Madhya Pradesh 0.534 0.319 0.215 0.297 0.438 Maharashtra 0.096 0.025 0.071 0.738 0.898 Orissa 0.317 0.135 0.182 0.509 0.650 Punjab 0.145 0.035 0.110 0.729 0.872 Rajasthan 0.728 0.579 0.149 0.166 0.150 Tamil Nadu 0.484 0.302 0.182 0.093 0.219 Uttar Pradesh 0.242 0.120 0.122 0.483 0.582 West Bengal 0.559 0.379 0.179 0.282 0.376 All India 0.325 0.210 0.115 0.428 0.463 Notes: Households classified into poor or non-poor based on poverty lines for Schedule 10 that would yield the

same state-specific poverty rates as estimated from Schedule 1, and reported in Table 1. All-India figures reported

in this table include only the states shown. Source: Authors‘ calculations from NSS (2009/10).

28

Table 5: Participation rates and targeting by caste

Scheduled

Tribes Scheduled

Castes

Other

Backward

Castes

Weighted

mean for

ST, SC

and OBC Others

Targeting

differential for

ST/ backward

castes Andhra Pradesh 0.567 0.434 0.382 0.412 0.150 0.262

Assam 0.192 0.179 0.163 0.174 0.191 -0.017

Bihar (0.087) 0.185 0.089 0.116 0.016 0.100

Chhattisgarh 0.519 0.435 0.504 0.500 0.214 0.286

Gujarat 0.340 0.289 0.180 0.252 0.070 0.181

Haryana (0.000) 0.105 0.044 0.071 0.018 0.054

Himachal Pradesh 0.392 0.413 0.294 0.376 0.298 0.077

Jammu and Kashmir (0.054) 0.134 0.109 0.114 0.090 0.024

Jharkhand 0.204 0.268 0.155 0.197 0.149 0.048

Karnataka 0.186 0.160 0.042 0.089 0.054 0.035

Kerala (0.168) 0.238 0.098 0.123 0.088 0.035

Madhya Pradesh 0.567 0.442 0.334 0.433 0.211 0.222

Maharashtra 0.063 0.017 0.074 0.058 0.015 0.044

Orissa 0.323 0.220 0.224 0.253 0.100 0.153

Punjab (0.000) 0.104 0.016 0.082 0.009 0.074

Rajasthan 0.816 0.654 0.581 0.644 0.444 0.200

Tamil Nadu (0.286) 0.523 0.279 0.338 0.069 0.269

Uttar Pradesh (0.140) 0.325 0.118 0.191 0.044 0.146

Uttarakhand 0.388 0.513 0.082 0.321 0.278 0.043

West Bengal 0.656 0.507 0.449 0.521 0.362 0.159

All India 0.415 0.336 0.214 0.279 0.155 0.124

Notes: TD for ST/backward castes is defined as the difference between the (weighted) mean participation rate of ST,

SC and OBCs and that for others. ST figures in parentheses had less than 100 sampled ST households and so might

be unreliable. All-India figures include states not shown in the table. Source: Authors‘ calculations from NSS

(2009/10).

29

Table 6: Average wages on MGNREGS and in casual labor 2009/10

Average wage rate

on MGNREGS

(Rs/day)

Average casual wage rate (Rs/day)

Overall Male Female

Andhra Pradesh 91.9 98.5 115.4 75.7

Assam 87.0 90.1 94.4 74.9

Bihar 97.5 79.4 81.0 65.8

Chhattisgarh 82.3 68.8 70.8 65.5

Gujarat 89.3 83.3 87.3 71.0

Haryana 150.9 139.6 146.1 99.1

Himachal Pradesh 109.5 139.6 141.4 110.2

Jammu & Kashmir 93.3 158.3 157.5 n.a.

Jharkhand 97.7 101.2 103.6 82.2

Karnataka 86.0 84.5 96.9 62.8

Kerala 120.6 206.5 226.6 119.3

Madhya Pradesh 83.7 69.0 74.5 58.1

Maharashtra 94.3 75.2 86.0 58.2

Orissa 105.9 75.6 81.0 59.1

Punjab 123.5 130.4 133.5 91.8

Rajasthan 87.4 125.7 132.3 94.3

Tamil Nadu 71.6 110.8 132.1 72.6

Uttar Pradesh 99.5 94.3 97.0 69.2

Uttarakhand 99.0 118.7 122.1 96.7

West Bengal 90.4 85.3 87.8 65.9

All India 90.2 93.1 101.5 68.9 Notes: MGNREGS wage rates estimated as total expenditure on wages (excluding skilled or semi-skilled) divided

by total number of person days of employment for FY 2009/10 (April 2009 to March 2010). Casual wages for June

2009 to July 2010 period, based on NSS 66th

round survey. ―All India‖ includes smaller states not reported. Note

that we do not report the female wage rate for Jammu & Kashmir, as we found it was based on a sample of only 7

observations and was not reliable. (The sample estimate of 206.5 was also implausibly high relative to the male

wage). Sources: Casual wages from Key Indicators of Employment and Unemployment in India, 2009/10, NSSO,

Govt. of India (June 2011). MGNREGS expenditure and employment data are from the state-wise Monthly

Progress Reports (www.nrega.nic.in).

30

Table 7: Regressions for the female share of employment in MGNREGS

Full

sample Sample with

headcount index

available

Sample with male and

female wages available

Constant 0.676 (8.671)

0.829 (5.341)

0.193 (0.887)

0.131 (0.246)

0.697 (3.922)

Rationing rate -0.505 (-3.758)

-0.419 (-2.898)

-0.307 (-2.904)

-0.456 (-5.108)

-0.469 (-5.972)

Headcount index of rural

poverty n.a. -0.515

(-1.538) 0.241

(1.518) n.a. n.a.

Female casual non-PW

wage (log) n.a. n.a. -0.549

(-2.056) -0.744

(-3.475) n.a.

Male casual non-PW wage

(log) n.a. n.a. 0.675

(5.156) 0.806

(3.556) n.a.

Female wage relative to

male wage (log) n.a. n.a. n.a. n.a. -0.836

(-3.301) R

2 0.336 0.467 0.811 0.692 0.687 SEE 0.164 0.145 0.093 0.111 0.108 N 20 18 18 19 19 Notes: The dependent variable is the share of total person days of employment on MGNREGS going to women. The

rationing rate is the share of those who wanted work who did not get it. The headcount index is the percent of

population below the poverty line. The t-ratios in parentheses are based on White standard errors.

31

References

Basu, Arnab K., 2011, ―Impact of Rural Employment Guarantee Schemes on Seasonal Labor

Markets: Optimum compensation and Workers‘ Welfare,‖ Journal of Economic

Inequality, in press.

Basu, Arnab K., Nancy H. Chau, and Ravi Kanbur, 2009, ―A Theory of Employment

Guarantees: Contestability, Credibility and Distributional Concerns,‖ Journal of Public

Economics 93: 482-497.

Besley, Timothy, and Stephen Coate, 1992, ―Workfare vs. Welfare: Incentive Arguments for

Work Requirements in Poverty Alleviation Programs,‖ American Economic Review

82(1): 249–61.

Datt, Gaurav and Martin Ravallion, 1994, ―Transfer Benefits from Public Works Employment,‖

Economic Journal 104: 1346-1369.

Drèze, Jean and Reetika Khera, 2009, ―The Battle for Employment Guarantee,‖ Frontline, 26(1):

3-16.

Dutta, Puja, Rinku Murgai, Martin Ravallion and Dominique van de Walle, 2012, Rozgar

Guarantee? Assessing India’s Biggest Anti-Poverty Program in India’s Poorest State,

World Bank, Washington DC, forthcoming.

Gaiha, Raghav, 1997, ―Rural Public Works and the Poor: The Case of the Employment

Guarantee Scheme in India,‖ in S. Polachek (ed.) Research in Labour Economics,

Connecticut: JAI Press.

Imbert, Clément and John Papp, 2011 ―Equilibrium Distributional Impacts of Government

Employment Programs: Evidence from India‘s Employment Guarantee,‖ mimeo, Paris

School of Economics.

Jha, Raghbendra, Sambit Bhattacharyya, Raghav Gaiha and Shylashri Shankar, 2009,

―‗Capture‘ of Anti-Poverty Programs: An Analysis of the National Rural Employment

Guarantee Program in India,‖ Journal of Asian Economics 20(4): 456-464.

Khera, Reetika, (ed.), 2011, The Battle for Employment Guarantee, New Delhi: Oxford

University Press.

Lanjouw, Peter and Martin Ravallion, 1999, ―Benefit Incidence and the Timing of Program

Capture,‖ World Bank Economic Review 13(2): 257-274.

Liu, Yanyan and Klaus Deininger, 2010, ―Poverty Impacts of India‘s National Rural

Employment Guarantee Scheme: Evidence from Andhra Pradesh, mimeo, World Bank.

Lokshin, Michael, 2006, ―Difference-Based Semiparametric Estimation of Partial Linear

Regression Models,‖ STATA Journal 6(3): 377-384.

Ravallion, Martin, 1990, ―Market Responses to Anti-Hunger Policies: Wages, Prices, and

Employment,‖ In Jean Drèze and Amartya Sen (eds) The Political Economy of Hunger:

Volume 2. Oxford: Oxford University Press.

______________, 1991, ―On the Coverage of Public Employment Schemes for Poverty

Alleviation,‖ Journal of Development Economics 34: 57-80.

32

______________, 2000, ―Monitoring Targeting Performance when Decentralized Allocations to

the Poor are Unobserved,‖ World Bank Economic Review 14(2): 331-45.

______________, 2009, ―How Relevant is Targeting to the Success of an Antipoverty

Program?‖ World Bank Research Observer 24(3): 205-231.

Ravallion, Martin, Gaurav Datt, and Shubham Chaudhuri, 1993, ―Does Maharashtra‘s

‗Employment Guarantee Scheme‘ Guarantee Employment? Effects of the 1988 Wage

Increase,‖ Economic Development and Cultural Change 41: 251–75.

Shariff, Abusaleh, 2009, ―Assessment of Outreach and Benefits of National Rural Employment

Guarantee Scheme in India,‖ Indian Journal of Labor Economics 52(2): 243-268.